In ops prometheus we limit the envoy metrics scraped via a relabel config:
# Envoy produces a ton of metrics, but for now we're just interested in # upstream and downstream requests latencies and counts, as well as connection # stats. So just keep those and nothing else. 'metric_relabel_configs' => [ { 'source_labels' => ['__name__'], 'regex' => '^envoy_(http_down|cluster_up)stream_(rq|cx).*$', 'action' => 'keep' }, ]
We should do the same for the k8s prometheus instances by default.